Ownership structure and risk in publicly held and privately owned banks

Transcription

1 Ownership structure and risk in publicly held and privately owned banks Thierno Amadou Barry, Laetitia Lepetit, Amine Tarazi To cite this version: Thierno Amadou Barry, Laetitia Lepetit, Amine Tarazi. Ownership structure and risk in publicly held and privately owned banks. Journal of Banking and Finance, Elsevier, 2011, 35, pp < /j.jbankfin >. <hal > HAL Id: hal Submitted on 30 Dec 2014 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

2 Ownership structure and risk in publicly held and privately owned banks Thierno Amadou Barry, Laetitia Lepetit and Amine Tarazi * Université de Limoges, LAPE, 5 rue Félix Eboué, Limoges Cedex, France This version : September 2010 Abstract Using detailed ownership data for a sample of European commercial banks, we analyze the link between ownership structure and risk in both privately owned and publicly held banks. We consider five categories of shareholders that are specific to our dataset. We find that ownership structure is significant in explaining risk differences but mainly for privately owned banks. A higher equity stake of either individuals/families or banking institutions is associated with a decrease in asset risk and default risk. In addition, institutional investors and non-financial companies impose the riskiest strategies when they hold higher stakes. For publicly held banks, changes in ownership structure do not affect risk taking. Market forces seem to align the risk-taking behavior of publicly held banks, such that ownership structure is no longer a determinant in explaining risk differences. However, higher stakes of banking institutions in publicly held banks are associated with lower credit and default risk. JEL Classification: G21; G32 Keywords: Ownership structure; Bank risk; European banks; Market discipline * Corresponding author: Tel: ; fax : addresses : thierno.barry@unilim.fr (T. Barry); laetitia.lepetit@unilim.fr (L. Lepetit); amine.tarazi@unilim.fr (A. Tarazi). 1

3 1. Introduction The past three decades have been characterized by repeated banking crises, such as the financial crisis of 2008, the U.S. savings and loans debacle of the 1980s, the Mexican crisis, and the 1997 Asian and 1998 Russian financial crises. Such episodes highlight the inherently unstable nature of banking and the tendency of banks toward excessive risk taking. In this paper, we focus on a driving force behind the risk-taking incentives of banks namely, shareholders behavior and their incentives to take higher risk. The issue of ownership structure is of particular interest for the banking industry because several factors interact with and alter governance, such as the quality of bank regulation and supervision and the opacity of bank assets. Moreover, banking systems have faced major changes during the past 20 years. With financial deregulation and market integration, the scope of banks activities has been completely reshaped, from traditional intermediation products to an array of new businesses. These trends have led to substantial consolidation in the banking industry and, consequently, to significant changes in ownership and capital structure. In addition, institutional ownership of common stock has increased substantially over the past 20 years, which also implies changes in corporate governance and banks behavior in terms of risk taking. However, because of greater separation of ownership and control, firms with publicly held equity face different agency problems than privately owned firms. 2 Indeed, in publicly held banks, ownership is more likely to be dispersed among a large number of shareholders. This implies that the separation between shareholders and managers is more effective for publicly held banks than for privately owned banks. Such separation between shareholders and managers can increase information asymmetry and therefore create divergence in incentives (Jensen and Meckling, 1976). Privately owned banks are usually characterized by 2 Publicly held banks are banks that are publicly quoted (i.e. banks whose stocks are traded on a stock exchange). Privately owned banks are all other banks. 2

4 less separation between owners and managers. The latter have a relatively larger equity stake, and therefore their incentives are more closely aligned with those of shareholders. Moreover, in privately owned banks, shareholders can more easily gain access to managers private information that facilitates the monitoring of their actions. The choice to be publicly held or privately owned also implies differences in terms of market discipline and access to capital markets. For publicly traded banks, market forces can influence risk-taking incentives. On the one hand, the market is expected to monitor or influence banks risk behavior, and therefore the impact of ownership changes on risk cannot be assessed without considering incentives driven by financial markets in terms of discipline (Bliss and Flannery, 2002; Flannery, 2001). In the Basel II Capital Accord, market discipline is one of the three pillars, along with capital regulation (first pillar) and banking supervision (second pillar). The idea is to rely on market forces to enhance banking supervision or to mitigate shareholders risk-taking incentives, and consequently market discipline should play an important role for publicly held banks and, to some extent, for privately owned banks that strongly rely on market debt. On the other hand, banks that are publicly held might have different objectives in terms of growth and risk-return strategies. Public equity is more liquid than private equity and thus can be raised at a lower cost. Therefore, if publicly held banks purpose to access capital markets is to finance faster growth opportunities, they are likely to take on more risk than privately owned banks. To our knowledge, no research has addressed whether risk may be different for privately owned banks and publicly held banks under specific ownership profiles. Working with a sample of U.S. bank holding companies (BHC), Kwan (2004) finds no difference in loan quality and earnings variability between traded BHCs and privately owned BHCs, whereas Nichols et al. (2009) find that publicly held banks exhibit relatively larger loan loss allowances and loan loss provisions than privately owned banks. Thus, our aim herein is to 3

5 assess banks risk-taking behavior by combining the two interrelated dimensions of ownership structure and market discipline. According to theoretical and empirical literature, agency problems and risk-taking behavior are different depending on the nature of the shareholder. A first issue is the conflict of interest between managers and shareholders identified by Jensen and Meckling (1976). Theory indicates that shareholders with a diversified portfolio are motivated to take more risk for a higher expected return whereas managers take less risk to protect their position and personal benefits and to preserve their acquired human capital (Galai and Masulis, 1976; Jensen and Meckling, 1976; Demsetz and Lehn, 1985; Esty, 1998). Empirically, Saunders et al. (1990) were the first to test the relationship between banks ownership structure and their risk-taking incentives. They find a positive relationship between managerial stock ownership (proportion of stock held by managers) and risk taking. Moreover, they find that banks controlled by shareholders take more risk than banks controlled by managers. In line with Saunders et al. (1990), several studies find a significant effect of ownership concentration on risk taking but without any consensus on the sign of such a relationship. That is, some studies find a negative relationship, whereas others obtain a U-shaped relationship (or inverse U shape) between ownership concentration and risk (Gorton and Rosen, 1995; Chen et al., 1998; Anderson and Fraser, 2000), which could be explained by managers entrenchment. Moreover, Sullivan and Spong (2007) show that stock ownership by hired managers is positively linked with bank risk, meaning that under certain conditions, hired managers operate their bank more closely in line with stockholder interests. Existing research also analyzes how the level of ownership concentration affects bank performance. The effects of ownership concentration on firm performance are theoretically complex and empirically ambiguous. Shleifer and Vishny (1986) and Aghion and Tirole (1997) show that a concentrated ownership may improve firms performance by increasing 4

6 monitoring and alleviating the free-rider problem in takeovers. Conversely, other theoretical works show that large shareholders may exercise control rights to create private benefits and sometimes to expropriate smaller investors (Shleifer and Vishny, 1997). Another potential cost of concentration may result if managerial initiative is repressed by excessive monitoring (Burkart et al., 1997). Working with the 10 largest publicly listed banks in 48 countries, Laeven and Levine (2009) find that banks with more powerful owners tend to take higher risks. In contrast, working with a panel of 50 countries, Shehzad et al. (2010) find that when ownership concentration is greater than 50%, the volume of non-performing loans decreases. Their results further reveal that when shareholder protection rights are weak, ownership concentration is beneficial for the bank. Another well-developed issue in the literature involves comparing the performance (profitability and asset quality) of state-owned banks with that of their private counterparts. Agency costs within government bureaucracy can result in weak managerial incentives and misallocation of resources. According to the agency cost view, managers exert less effort than their private counterparts or divert resources for personal benefits, such as career concerns. From the political view of state ownership, government-owned banks are inefficient because of politicians deliberate policy of transferring resources to their supporters (Shleifer and Vishny, 1986; Shleifer, 1998). According to prior research, state-owned banks have poorer loan quality and higher default risk than privately owned banks (Berger et al., 2005; Iannotta et al., 2007). Iannota et al. (2007) also suggest that mutual banks and government-owned banks are less profitable than privately owned banks. Moreover, they find that governmentowned banks have poorer loan quality and higher default risk, whereas mutual banks have better loan quality and lower asset risk than both privately owned and government-owned banks. In addition, some research has shown that foreign-owned banks exhibit better 5

7 performance than other banks, particularly in developing countries (Claessens et al., 2001; Bonin et al., 2005; Micco et al., 2007). In addition to the issues of the manager/owner conflict and the differences between state-owned and privately owned firms, other aspects have been well established in the literature on non-financial firms but not on financial firms. First, institutional investors (e.g., investment companies, investment advisors, pension funds) who exercise significant voting power can shape the nature of corporate risk taking. In terms of shareholder size and expertise in processing information and monitoring managers, such investors are different from atomistic individual investors because they can exert greater control for reasons of economies of scale in corporate supervision. Pound (1988) shows that institutional investors can exercise control at a lower cost because they have more experience. However, managers and institutional investors may also form an alliance, in which insider interests take priority over the maximization of firm value. At the same time, because institutional investors have a diversified portfolio of investments, they may have lower incentives to exercise control. Empirical evidence (Acker and Athanassakos, 2003) based on non-financial firms does not provide conclusive results on the effect of control by institutional investors on firm value. Second, family-owned firms are perceived not only as less willing to take risk but also as less profitable. More generally, firms with large, undiversified owners such as founding families may forgo maximum profits because their wealth is not sufficiently diversified. Families also limit executive management positions to family members, suggesting a restricted labor pool from which to obtain qualified and capable talent, potentially leading to competitive disadvantages relative to non-family-owned firms (Morck et al., 2000). However, James (1999) posits that families have longer investment horizons that lead to greater investment efficiency. Stein (1988, 1989) shows that the presence of shareholders with relatively long investment horizons can mitigate the incentives for myopic investment decisions by 6

8 managers. Regarding the banking industry, little research has analyzed this issue. Laeven (1999) considers different forms of bank ownership, including state-owned, foreign-owned, company-owned and family-owned banks, but not banks owned by institutional investors. Working with a panel of Asian banks before the Asian crisis of 1997, he finds that familyowned banks were among the most risky banks, along with company-owned banks, whereas foreign-owned banks took little risk relative to other banks. The objective of this paper is to extend the current literature regarding how ownership structure affects bank risk taking and profitability in several directions. First, we work on a broader classification of shareholders by considering not only the equity held by managers, individuals/families, and non-financial companies but also the equity held by institutional investors and banks. Second, we consider the proportion of equity held by each category of owner instead of using dummy variables to divide ownership into mutually exclusive categories, as does most of the previous studies on bank ownership (Berger et al., 2005; Bonin et al., 2005; Boubakri et al., 2005; Williams and Nguyen, 2005). Therefore, we can measure the level of ownership dispersion/concentration within each of the five categories of shareholders we consider. In addition, we can check whether the level of ownership dispersion matters when assessing the relationship between ownership structure and bank risk/profitability (Laeven and Levine, 2009; Shehzad et al., 2010) by examining ownership dispersion within each category of shareholders. Working with continuous variables instead of binary variables also enables us to analyze how the interaction of equity held by different types of shareholders influences banks risk-taking behavior. This allows us to study the link between ownership structure and risk more thoroughly by dealing with the issue of possible coalitions among different categories or groups. Nevertheless, for consistency with previous studies we also study the link between risk and the nature of the main shareholder. Third, by investigating the link between ownership structure and risk for both listed (publicly held) and 7

9 non-listed (privately owned) banks, we question the ability of market forces to influence bank risk-taking behavior (market discipline) under different ownership arrangements. Fourth, previous studies that use a detailed breakdown of the stakes held by different categories of owners were mostly dedicated to U.S. banks and therefore did not consider as many categories of shareholders because ownership of banks by non-financial companies is not permitted. By studying European banks, we are able to introduce an additional category, nonfinancial firms, which according to the literature are controversial in influencing the management of financial institutions. Studies on European banks have focused on the nature of ownership (e.g., public, private, mutual, cooperative) rather than on the structure of ownership in privately owned banks. In this paper, we consider only one category of banks. We focus on commercial banks because they have homogeneous objective functions; to our knowledge, this is the first study to explore the relationship between ownership structure and risk for European commercial banks. We work on a panel of European banks through the period. Our results show that different ownership structures imply different levels of risk and profitability, but such findings hold mainly for privately owned banks. Publicly held banks with different ownership structures do not present different levels of risk and profitability, suggesting that market forces align the risk behavior of such banks. The remainder of the paper is structured as follows: In Section 2, we describe the data and variables. In Section 3, we present the methodology and the hypotheses. In Section 4, we discuss the empirical results. In Section 5, we report robustness checks and discuss further issues. Section 6 concludes the paper. 2. Data, variables and descriptive statistics 2.1. Data collection and sample definition 8

10 We take the annual data used in this paper from BankScope Fitch IBCA, which provides information on financial statements and ownership structure for financial institutions worldwide. We collect the percentage of stocks held by shareholders by considering the following categories: managers/directors, institutional investors, non-financial companies, self-ownership, individual/family investors, banks, foundations/research institutes, government, unnamed private shareholders and other unnamed shareholders. BankScope Fitch IBCA also provides for listed banks data on the percentage of stocks held by the public (i.e., by small, not identified investors). We use a sample consisting of an unbalanced panel of annual report data from 1999 to 2005 for a set of European commercial banks established in 16 Western European countries: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, Switzerland and the United Kingdom. 3 As argued previously, we do not consider other types of banks (e.g., cooperative, mutual) to ensure that all the banks in our sample follow the same profit maximization objective (homogeneous objective function). We identify in BankScope 1586 commercial banks for which income statements and balance sheets are provided for the period. 4 We delete all the banks with less than five consecutive years of time series observations, 5 which leaves us with 688 banks. Of these banks, we isolate 320 for which detailed data on direct ownership are available for the years 2001, 2003 and 2005 in the annual financial statement. 6 Eventually, we apply other selection criteria and end up with a 3 We exclude Norway from our analysis because no banks provide data consistent with the criteria we use to build and clean our database. 4 All the banks in our sample publish their annual financial statements at the end of the calendar year. We consider local Generally Accepted Accounting Principles (GAAP) for all our sample period. 5 This condition enables us to accurately compute the standard deviations of some variables to define risk indicators. 6 Each annual financial statement provides information on the banks ownership structure for the current year and the previous two years. For example, the report of the year 2001 gives information on the ownership structure of the years 1999, 2000 and In our study, we consider the direct owner, who can be different from the ultimate owner (e.g., 20% of a bank s stocks can be owned by a firm [direct owner] in which a family might have a stake of 10%). We use direct ownership to consider the different categories that directly exert control. We do not consider the ultimate owners because BankScope only provides information on such owners since 2004 and only for shareholders with stakes greater than 25%. 9

11 smaller sample of banks. First, we only consider banks with a stable ownership structure by comparing the proportion of equity held by the main shareholders over the period. This restriction is important to accurately analyze the impact of ownership structure on the performance and risk of banks. Because our aim is to focus on the influence of different categories of shareholders on management, we need to exclude short-term ownership and hitand-run strategies that do not shape management behavior and, therefore, bank risk/profitability in a given direction. 7 Thus, we only keep banks for which the ownership shares of the main category of shareholders fluctuate by less than 10% over the considered period. In total, 249 banks are consistent with this criterion, which enables us to work on a firm-level homogeneous sample. The final sample consists of 249 European commercial banks, 80 of which are listed as publicly traded banks 8 (see Table 1 for further details on the distribution of banks by country). Of these banks, 191 have a major shareholder with a stake greater than 50% throughout the whole sample period, and 58 banks (of which 44 are listed) exhibit ownership shares by the main shareholder fluctuating by less than 10%. We also consider a subsample that satisfies the criteria that the sum of the different shares displayed in BankScope is at least equal to 99%. 9 This criterion leaves us with 198 banks, 29 of which are listed. We test the robustness of our results by running our estimations on both the large sample of 249 banks and the restricted sample of 198 banks. To be consistent with previous 7 As an anonymous referee noted, this restriction could be relaxed to check for robustness of results and to possibly capture deeper insights in our investigation. Indeed, even large changes in ownership structure may not alter management behavior, specifically for publicly and widely held institutions. However, to estimate the risk measures, we need to compute standard deviations over the whole sample period because we use annual data that cover a relatively short period ( ). In other words, we can assign only one risk value to a bank throughout the sample period. Relaxing this restriction by considering banks that had ownership changes would not allow us to track possible risk changes. 8 Our full dataset contains 137 listed banks. We need to delete (1) 7 banks with less than five years of time series observations; (2) 31 banks for which ownership is not detailed in the three reports provided for the years 2001, 2003 and 2005; and (3) 19 banks that exhibit a change in ownership structure between 1999 and The banks classified as listed were continuously publicly quoted throughout the sample period. Seven banks that were delisted in the early years of our sample are classified as non-listed banks in our study. We do not include banks that were non-listed (listed) for some years but became listed (delisted). Our approach, which considers a sample of banks with stable ownership structures, might explain why few banks were alternatively listed and non-listed on a stock exchange over the sample period. 9 The data on ownership structure provided by BankScope (% share of each type of owner) do not always add up to 100%, particularly for listed banks, because we do not always have the percentage held by the public. 10

12 studies, we also conduct estimations on the sample of 191 banks for which we have a major shareholder with a stake greater than 50%. Insert Table 1 here Table 2 presents descriptive statistics for both our sample of 249 banks and the largest sample of 1586 commercial banks available in BankScope Fitch IBCA for our period of analysis. We use data from consolidated accounts if available and from unconsolidated accounts otherwise. Insert Table 2 here 2.2. Risk variables We consider different measures of asset risk and default risk commonly used in the literature. We compute three standard measures of risk for each bank throughout the period under study on the basis of annual accounting data: the standard deviation of the return on average assets (SDROA), the standard deviation of the return on average equity (SDROE), 10 and the mean of the ratio of loan loss provisions to net loans (M_LLP). We also compute default risk measures. First, we use the Z-score proposed by Boyd and Graham (1986), which indicates the probability of failure of a given bank (Z). 11 Higher values of Z-scores imply lower probabilities of failure. Second, we use the ZP Score (ZP) as in Goyeau and Tarazi (1992) and Lepetit et al. (2008) and its two additive components (ZP1 and ZP2). 12 ZP1 is a measure of bank portfolio risk, and ZP2 is a measure of leverage risk. Table 2 provides statistics for our measures of asset risk and default risk, on average for the whole sample of banks, and for the panel of non-listed and listed banks. Mean tests show that there are no significant differences in risk between our two samples of publicly owned and privately owned banks. These results are consistent with those of Kwan (2004), 10 We define average equity and average assets at time t as (amount outstanding at time t + amount outstanding at time t 1)/2. 11 Z (100 average ROE) / SDROE, where ROE and SDROE are expressed in percentages. 12 average average(totalequities / Totalassets) ZP = ZP 1 + ZP 2 = ROA. SDROA SDROA 11

13 who examines a panel of U.S. banks. However, unlike his findings, our sample of European publicly held banks exhibits, on average, a higher profitability than our sample of European privately owned banks. An explanation for the higher profitability for listed banks could be that such banks can raise additional equity capital at lower transaction costs, which enables them to generate faster growth in equity and assets and, thus, to become larger. These banks might also benefit from economies of scale and generate higher profit per unit of risk than privately owned banks Ownership variables In our study, we code the ownership structure according to the stockholder information contained in the BankScope database. Because our aim is to analyze how the interaction of equity held by different types of shareholders influences banks risk-taking behavior, we must consider as many categories of owners as possible. However, we only keep the categories of owners for which we are able to identify their nature, behavior and incentives to take risk. Therefore, we exclude three categories of owners that BankScope provides: public, unnamed private shareholders and other unnamed shareholders. We also require each category of owner to hold equity in at least five banks. These criteria lead us to exclude three categories of owners: self-owned, foundation and government. 13 Consequently, we end up with five categories of owners in our study: (1) managers/directors (MANAGER), (2) non-financial companies (COMPANY), (3) individual/family investors (FAMILY), (4) banks (BANK), and (5) institutional investors (INSTITUT), such as insurance companies, financial companies and mutual and pension funds. We create five variables that report the proportion of equity held by each category of owner for each bank in our sample. This approach allows us to measure the dispersion of ownership and also to analyze the influence of different combinations of shareholders on bank 13 Few European banks have equity held by governments, and those that do are mostly German cooperative banks, which we do not consider in our sample. 12

14 risk and profitability. It also enables us to account for possible coalitions among different categories of shareholders. 14 Descriptive statistics of our sample of 249 European commercial banks show that the major shareholder category is other banking institutions with an average of 81.52% of equity. 15 Non-financial companies and institutional investors are also strongly involved in our sample banks because they hold equity in 78 and 55 banks of the 249 banks of our sample, respectively. Non-financial companies hold, on average, a higher percentage of equity (39.48%) than institutional investors (35.40%). Individuals/families are involved in a relatively few number of listed and non-listed banks (25 banks). The category managers/directors holds equity in only 8 banks, of which 7 are listed banks, and the average proportion of stocks they hold is low (9.51%) compared with the other types of owners. All the shareholder categories exhibit, on average, higher stakes in non-listed banks than in listed banks. Statistics further show that the proportions of equity of each category of owner (except managers/directors) are well distributed in the interval [0 100]. There is also strong heterogeneity among different types of shareholders regarding risk and default risk indicators, which enables us to analyze the behavior of banks according to their ownership structure. 14 As an anonymous referee noted, considering continuous variables (proportion of equity held by each category of shareholders) instead of dummy variables to capture the nature of the main shareholder requires significant attention. Indeed, changes in ownership from 3% to 5% will not have the same impact as changes from 49% to 51%. In our setting, we consider the proportion of equity held by several categories of shareholders that adds up to 100%. Therefore, a change in ownership of one category cannot be assessed without considering the outcomes for all the other categories. Specifically, an increase from 3% to 5% of one category can either significantly reduce the influence of another category (a drop from 51% to 49%) or not. Similarly, an increase from 24% to 26% might or might not have a significant impact depending on the stakes held by the other categories. Therefore, the impact of a change in ownership will depend on how equity is initially distributed among the five shareholder categories and not only on the initial proportion held by the category itself. As a whole, we cannot presume that a change in ownership from 4% to 6% will have a weaker impact than a change from 24% to 26%. The advantage of our approach is that it captures the effect on risk of various combinations of equity held by the different categories of shareholders. It also allows us to measure the level of dispersion/concentration within each of the five categories. However, because of the complexity of our approach, which simultaneously considers the stakes of all the shareholder categories, we also consider the nature of the main shareholder by introducing dummy variables instead of continuous variables. 15 Extensive tables on the descriptive statistics of banks ownership structure are available on request. We also compare our sample of 249 commercial banks with the larger population of European commercial banks for which BankScope Fitch IBCA provides information on the ownership structure in 2005 (905 commercial banks) by examining possible differences. The frequencies of banks for which each category of owner holds a positive percentage of equity in our sample are not significantly different from those of the largest sample of 905 banks. 13

15 We also compare asset risk, default risk, profitability and asset growth of publicly held and privately owned banks when held by the same main category of shareholders 16 (see Figure 1). Figure 1 shows differences in asset risk and profitability between listed and nonlisted banks for a given shareholder type, but the differences are not statistically significant. However, we find that publicly held banks exhibit higher average asset growth rates than privately owned banks, regardless of the category of the main shareholder. These results suggest that, as discussed previously, publicly held banks can raise equity more easily and at lower cost to generate faster growth. Insert Figure 1 here We further measure the ownership dispersion/concentration of our sample of European commercials banks to analyze its possible impact on the risk-taking behavior of banks as in Laeven and Levine (2009) and Shehzad et al. (2010). Our data provide the proportions of total equity held by different categories of owners but not the stakes held by each investor at the individual level. Therefore, we need to check whether the level of ownership dispersion in each of the five categories of shareholders matters when assessing the relationship between ownership structure and bank risk/profitability. For this purpose, we measure ownership dispersion/concentration by computing a Herfindahl index for each of the five categories of shareholders 17 (HERF_MANAGER, HERF_FAMILY, HERF_INSTITUT, HERF_COMPANY and HERF_BANK). Descriptive statistics show that, on average, ownership is relatively well concentrated for all the shareholder categories. This is consistent with the studies of La Porta et al. (1999) and Becht and Roell (1999), who show that the 16 The main category of shareholders is the one with the highest level of equity holding. 17 For example, for the category INSTITUT, we compute for each bank i the variable OS j, defined by the ratio of the percentage of equity held by each institutional investor j to the total percentage of equity held by all the institutional investors. We then compute the Herfindahl index as n 2 OS j, where j represents the category of shareholders INSTITUT and n is the total number of institutional investors that hold equity in the bank i. For example, if bank i has two institutional investors holding 10% of total equity and 45% of total equity, the Herfindahl index will take the value of 0.70, indicating a relatively high level of concentration for the category INSTITUT. j=1 14

16 ownership structure of firms around the world presents a relatively high degree of ownership concentration. The category BANK exhibits a relatively high level of ownership concentration, followed by FAMILY and INSTITUT. The category with the highest number of shareholders per bank is COMPANY, with an average of 4.25 shareholders and a maximum of 66 shareholders involved in the same bank. We also find no significant differences in terms of risk and profitability between banks with a relatively high level of concentration for the categories of FAMILY, INSTITUT and COMPANY and those with a relatively low level of ownership concentration in these categories. In addition, there seems to be no impact of ownership concentration/dispersion on banks risk-taking behavior. However, when banks hold equity in other banks and their stakes are concentrated, we find that asset risk and profitability are significantly higher than in banks with a more dispersed bank ownership. 3. Method and hypotheses tested Our first objective is to analyze whether commercial banks with different ownership structures present significant differences in risk and profitability. We also investigate whether market discipline can influence the relationship between ownership structure and risk. Therefore, we test two hypotheses by considering two specifications. Hypothesis 1: Different ownership structures imply different levels of risk and profitability. We use the following econometric model to test hypothesis 1: Model 1 Y MANAGER FAMILY COMPANY BANK LISTED i 0 1 i 2 i 3 i 4 i 5 i M_LNTA M_OEQUITY M_DEPOSIT M _ CIR 6 i 7 i 8 i 9 LAMBDA LAMBDA*LISTED COUNTRY 10 i 11 i j j i j 1 15, 15

17 where Y i is a measure of asset risk (SDROA, SDROE and M_LLP), default risk (Z, ZP, ZP1 and ZP2) or profitability (the mean of the return on average assets, M_ROA, and the mean of the return on average equity, M_ROE) 18 ; MANAGER, FAMILY, COMPANY and BANK represent the percentage of stock held by managers/directors, individuals/families, nonfinancial companies, and banks, respectively 19 ; LISTED is a dummy variable that takes the value of 1 if the bank is listed on the stock market and 0 if otherwise; M_LNTA is the mean of the natural logarithm of total assets; M_OEQUITY is the mean of the ratio of equity to total assets orthogonalized with total assets; M_DEPOSIT is the mean of the ratio of deposits to total assets; M_CIR is the mean of the ratio of total operating expenses to total operating income; LAMBDA is the inverse Mills ratio estimated for each bank from the first-stage probit model; and COUNTRY is a country dummy variable. We consider five categories of owners that may influence banks risk-taking behavior (MANAGER, FAMILY, COMPANY, BANK and INSTITUT). In our specification, we remove INSTITUT from Model 1 to use institutional investors as a benchmark ownership share. The theoretical link between risk and institutional ownership is the most settled. Because institutional investors hold shares in sufficiently diversified investment portfolios, they are assumed to favor all positive net present value investments at the individual bank level. As shareholders, institutional investors are indifferent to the riskiness of an investment in a specific bank and are only concerned about expected return. By removing INSTITUT from our set of independent variables, we can analyze, with Model 1, whether a shift in ownership from institutional investors to another category of owners results in an increase or a decrease in risk 18 We do not include an independent variable reflecting asset risk when we consider profitability as the dependent variable, because we have a high degree of colinearity between our ownership variables and asset risk. 19 To account for non-linearity in the impact of ownership changes on risk, an alternative method would be to introduce the squared term of ownership variables in the regression. However, because the ownership variables range from 0% to 100%, they are highly correlated with their squared terms. To check for robustness, we also run regressions with the squared terms only. Introducing the square of the ownership variables instead of the ownership variables in level provides identical results. 16

18 and profitability. 20 The excluded shareholder group INSTITUT is the benchmark against which we evaluate the signs and the magnitudes of the coefficients on the four other ownership shares. The theory regarding the attitude of individuals/families toward risk stipulates that their portfolio is less diversified than those of other shareholders, particularly institutional investors, and therefore they have incentives to take less risk because, if the bank fails, they lose more than other shareholders. We expect that a shift in equity from institutional investors (INSTITUT) to individuals/families (FAMILY) results in a decrease in risk 21 ( 2 negative). Previous studies that analyzed the incentives of managers/directors to take risk were mostly dedicated to U.S. firms. Most studies have shown that when a manager/director holds a small share of the bank s equity, he/she may have incentives to take less risk. If the bank fails, he/she loses both his/her reputation and human capital investment. Our MANAGER variable is close to the proxy used by Saunders et al. (1990), which they compute as the number of shares held by executive and directors divided by the total number of shares outstanding. Note that the underlying assumption in the literature is that a low proportion of stocks held by managers is associated with a low share of the bank s stocks in the managers non-human wealth. In addition, a greater proportion of stocks held by managers is assumed to align their interest with those of shareholders as long as the larger investment in the bank s stocks does not prevent them from holding diversified portfolios. In our study, we do not have 5 20 ' ' Model 1 is defined as Y C Z, where C ij represent the five categories of shareholders and i 0 j ji 6 i i j=1 Z i is a vector of control variables. Because we have C =100- C 5i 4 j=1 ji, we can rewrite Model 1 as 4 ' ' ' ' follows: Y ( 100 ) ( )C Z. We can then estimate the following Model: i 0 5 j 5 ji 6 i i j=1 4 Y C Z, with i 0 j ji 6 i i j=1 ' ' ' ' = α +100α and =α -α, j=1,.., j j 5 21 We give here the expected sign for the measures of asset risk (SDROA, SDROE and M_LLP). We expect the opposite sign for the default risk measures (Z and ZP) because a lower Z-score value implies a higher probability of failure. 17

19 information about managers wealth and the level of diversification of their investment portfolio. We assume that the managers portfolios are less diversified than those of our benchmark, institutional investors. Therefore, we expect a negative coefficient for the variable MANAGER ( 1 negative). We also consider shares held by non-financial companies (COMPANY). Banks with a large portion of stocks held by firms are prone to increase the riskiness of loans granted to owners. Moreover, if a bank is behind an industrial group, the group management will have incentives to manipulate the bank to maximize the wealth of ultimate owners. Therefore, banks that are controlled by firms might have incentives to encourage riskier strategies than other categories, such as individuals/families. In addition, it could be argued that nonfinancial companies might hold sufficiently diversified asset portfolios similar to institutional investors. If this is the case, their risk incentives could be aligned with those of institutional investors. However, our data do not provide information on the structure of their investment portfolios. Therefore, the impact of a shift in equity from institutional investors to nonfinancial companies on bank risk is undetermined ( 3 non-significant or positive/negative). Banks (BANK) represent the fourth category of shareholders. Our statistics show that banks hold important stakes in other banks. When a bank owns another bank, the important risk return relationship and strategies are expected to be handled by the parent company, and not at its subsidiary firm. However, banks as a shareholder might encourage relatively conservative risk-taking strategies at the individual bank level for both safety net reasons and reputation concerns. In the event of financial distress or failure, the parent bank is expected to support its subsidiary, which can be costly. Therefore, we expect a negative coefficient for the variable BANK ( 4 negative). The control variables we introduce account for size differences (mean of the natural logarithm of total assets M_LNTA), business differences (mean of the ratio deposits to total 18

20 assets M_DEP), leverage differences (mean of the ratio equity to total assets M_EQUITY) and managerial efficiency differences (mean of the ratio total operating expenses to total operating income, M_CIR). We also use alternative control variables (the ratio of loans to total assets and the ratio of net non-interest income to net operating income) to check for robustness. Both M_LNTA and M_EQUITY are highly correlated, and thus the leverage ratio is orthogonalized with total assets (M_OEQUITY). Because the information on the ownership structure of our sample of banks is invariant through time ( period) and because we compute our measure of asset risk and default risk using the standard deviations of ROA and ROE, we conduct cross-section regressions. Therefore, we compute the means of our three control variables over the whole sample period. We also control for possible country-specific effects by including country dummies (COUNTRY). We further check whether publicly held banks behave differently than privately owned banks, by including in Model 1 a dummy variable, LISTED, which takes the value of 1 if the bank is listed on the stock market and 0 if otherwise. We expect this dummy variable to capture differences in risk and profitability for listed and non-listed banks. Market exposure should influence the behavior of publicly held firms. However, the effect of market exposure on risk is unclear. On the one hand, market discipline should impose strong incentives on banks to conduct their business in a safe, sound and efficient manner, including an incentive to maintain a high level of equity capital to face potential future losses. On the other hand, publicly held banks can have access to additional equity at a lower cost than privately owned banks. Thus, publicly held banks might have a greater ability to become larger and make acquisitions. They also have a higher degree of freedom to manage their equity and meet the regulatory capital requirement, which gives them more flexibility to invest in risky projects with a higher expected return. Market forces might also impose a higher risk-adjusted return 19

21 for publicly held banks. Therefore, the expected sign associated with the variable LISTED is undetermined ( 5 non-significant or positive/negative). We potentially have two endogeneity problems in our regressions, one with our ownership variables, which are continuous, and one with the binary variable LISTED. We address these two problems separately. Some studies (Demsetz and Lehn, 1985; Himmelberg et al., 1999; Gugler and Weigland, 2003) suggest that ownership is endogenous because it is influenced by the firm s level of performance and risk. One can also argue that investors might be attracted by banks with different risk levels. Some investors might simply choose to invest in banks with higher risk profiles to maximize their utility. We test for the presence of an endogeneity bias in the estimated equation for the three ownership variables for which we might encounter such a problem (COMPANY, BANK and INSTITUT). We consider several instrumental variables related to the legal environment of the banking system and to the nature of the bank s activities. 22 We first check whether our instrumental variables are valid. We select the instrumental variables that are correlated with the variables that might be endogenous (COMPANY, BANK and INSTITUT). We then ensure that our instrumental variables are not redundant when we have at least two available instruments and that the instrumental variables are exogenous by using the Anderson likelihood ratio test and the Sargan test. Finally, we use the Hausman test to determine whether our ownership variables are endogenous. The Hausman tests show that the endogeneity problem is not a major issue, 23 which implies that ordinary least squares (OLS) should be an efficient estimator. 22 We consider several instrumental variables. First, we differentiate the banking systems according to their legal environment. We use the database of La Porta et al. (1998), which groups the countries into four general legal families: English common law origin, countries of French civil law origin, countries of German civil law origin and countries of Scandinavian civil law origin. Second, we classify our sample banks according to the nature of their activities (e.g., proportion of subsidiaries abroad, focus or diversification, extent of loan activities in the balance sheet). The strategies banks pursue do not change much over time and might influence the choice of shareholders. 23 The Hausman tests show that the null hypothesis of exogeneity is not rejected for INSTITUT (except for SDROA, Z and ZP2), COMPANY (except for SDROA) or BANK (except for ZP1 and ROE). 20

22 Whether a bank becomes publicly held or remains privately owned also raises potential endogeneity issues in our econometric specification. Banks will choose to become listed on a stock market or to remain privately owned on the basis of the expected future changes in growth and profitability. We account for possible endogeneity of this choice by using the Heckman (1979) two-stage approach as in Givoly et al. (2010) and Nichols et al. (2009). In a first stage, we use a probit model to determine the variables that influence the bank s choice to be publicly held or privately owned. We then use the estimates of the probit model to compute the inverse Mills ratio for each sample bank (LAMBDA). 24 In the second stage, we introduce the inverse Mills ratio as a control variable in Model 1. By including LAMBDA in Model 1, we control for the correlation between LISTED and the second-stage errors to obtain consistent coefficient estimates. We also introduce an interaction variable combining LISTED and LAMBDA to allow the coefficient to vary between listed and nonlisted banks. Our second objective in this paper is to further investigate the issue of market discipline. We test the extent to which market forces influence the behavior of publicly held banks under different ownership structures. As discussed previously, we can expect two effects from market discipline on the behavior of publicly held banks: (1) a decrease in risk, if market forces moderate the incentives of banks dominated by institutional investors or other shareholder categories that are rationally inclined to take higher risks, and (2) an increase in risk, if market forces align the objectives of publicly held banks to generate faster growth and obtain a higher risk-adjusted return. Because market forces might align the objective of listed banks, we expect that their ownership structure will not affect their risk level. These opposite 24 We use mean tests to compare balance sheet and income statement characteristics between listed and nonlisted banks (as in Table 2). We retain 10 variables of the 24 initially examined variables that are significantly different between listed and non-listed banks: consumer loans/total assets, total earning assets/total assets, total deposits/total assets, cash/total assets, ROA, liquid assets/total assets, net loans/total assets, asset growth rate, market funding/total assets, and equity/total assets. We use these variables to model the selection of public versus private status. The results from the probit model are available on request. The pseudo-r-square statistic indicates that the model explains almost 68% of the cross-sectional variation in the choice between public and private status in our sample. 21